Eocene Volcanic Complex from Central British Columbia: The Role of Fractional Crystallization during the Magmatic Evolution

Magmatism is the main contributor to the growth of the earth’s crust (e.g., [1–3]), and thus, the generation and modification of primitive melts is important for an understanding of the evolution of the continental crust. After anatexis, primitive melts can be emplaced on the surface or solidify at depth. A comprehensive statistical analysis of the chemical compositions and relationship between volcanic and plutonic rocks (e.g., [4]) show that in a subduction-related setting differentiation trends from mafic to felsic compositions are indistinguishable between plutonic and volcanic suites and that fractional crystallization is the dominant process for the generation of intermediate and felsic magmas. On the other hand, in continental rift environments, the fractionation of plutonic and volcanic suites differs and is more complex [4].

Crystallization of mineral phases from melts depends on physical parameters including pressure, temperature, and oxygen fugacity as well on the melt composition. The magma differentiation of tholeiitic and calc-alkaline basalts and related rocks has been extensively investigated by experimental studies (e.g., [5–7]). However, there is lack of information on fractional crystallization processes of rocks with subduction geochemical imprints that are generated at a postcollisional rift setting (e.g., [8, 9]). Moreover, there is debate regarding the suitability of the term calc-alkaline as the classification parameters are not uniformly applied [10–12]. The primary factors that control the evolution or development of intermediate to silicic “calc-alkaline” rocks are the timing of phase saturation of Fe-Ti oxide minerals, plagioclase, and mafic silicate minerals which is influenced by magmatic pressure, oxygen fugacity, and water contents [11].

The purpose of this paper is to evaluate the fractionation of consanguineous Eocene mafic to intermediate postcollisional volcanic/plutonic suites from central British Columbia. We compare whole-rock geochemical data as well as mineral chemistry of the rocks of the Buck Creek complex, a part of an Early to Middle Eocene belt of volcanic and plutonic rocks, which extends from central British Columbia to the northwestern United States.

The background geological and geochemical information on the complex was initially published by Church and Barakso [13] and Dostal et al. [14, 15]. The Eocene volcanic rocks form a prominent discontinuous belt in central and southern British Columbia, which runs inland, parallel to the coast at a distance of about 300 km and is more than 200 km wide (Figure 1). The volcanic chain (Challis-Kamloops belt of Souther [16]) extends for about 2,000 km from central British Columbia to Idaho and Wyoming in the northwestern United States. In British Columbia, the Eocene volcanic successions of the belt are voluminous and spread over an $area>65,000km2$ [17] stretching from central British Columbia (~55° N) to the United States-Canada border (49° N). The large volume of volcanic products together with the short duration of volcanism (~10 My) suggest that they were related to regional tectonic process. Traditionally, the Eocene magmatism has been associated with an eastward subduction of oceanic plates of the Pacific under North America. However, plate reorganization in the Pacific Ocean during the Paleocene to Eocene induced an abrupt change in plate convergence, which led to crustal extension [18–20], formation of numerous grabens, and block-faulted basins associated with the emplacement of major metamorphic core complexes and probably triggered decompression melting which produced the Eocene volcanic rocks during regional transtension in the Cordilleran orogeny (e.g., [17]).

The volcanic belt of central and south-central British Columbia is composed of isolated complexes that were emplaced typically between 55 and 45 Ma during a time of normal and dextral strike-slip faulting. The complexes were emplaced mostly in block-faulted basins and grabens although some of them form stratovolcanos, domes, and calderas. Stratigraphy of the individual complexes is unique and thus difficult to regionally correlate (e.g., [17, 21, 22]). Previous studies considered the volcanic rocks to be primarily calc-alkaline (i.e., medium-Fe to low-Fe) with an exception of the southern part where the belt also contains alkaline rocks. The volcanic rocks are associated with coeval plutonic rocks (e.g., [23–25]). These Eocene volcanic complexes are overstep sequences overlying a basement that was produced by Mesozoic amalgamation of pericratonic allochthonous terranes mainly of late Paleozoic and Mesozoic age and ancient cratonic North America [23, 26].

The Buck Creek volcanic complex in the interior of central British Columbia (Figure 1) is located at the northern part of the Nechako plateau, north of Francois Lake [13], near the town of Houston. The Nechako plateau hosts several Eocene volcanic suites (e.g., Ootsa Lake and Endako groups) compositionally similar to those of the Buck Creek complex (e.g., [17, 24]). The Buck Creek complex (BCC) is hosted by a block-faulted basin that occupies a ~3,000 km² area (Figure 2). The basin is filled by volcanic, volcaniclastic, and subordinate sedimentary rocks. At the base of the Buck Creek basin, the Tertiary fill started with a conglomerate unit, the Burns Lake Formation, which is 50-100 m thick, unconformably overlies the Mesozoic basement and is in turn overlain by the volcanic units. The Eocene volcanic rocks extruded during two short-lived (<1-2 My) periods, separated by a brief hiatus of no more than 1-2 My [14, 15]. The first period dated at $52±1Ma$ (Ar/Ar-whole-rock and Ar/Ar-biotite; [15]) produced high-K subalkaline volcanic rocks of the Goosly Lake Formation (GLV), which form a ~500 m thick sequence composed of 2-6 m thick subaerial, amygdaloidal mafic, and intermediate lava flows locally interbedded with volcanic breccia (Figure 3). Age dates from the base and top of the GLV are overlapping, pointing to the relatively fast eruption of the volcanic sequence [15]. The volcanic suite was intruded by several plugs that range from 0.5 to 2 km in diameter composed of porphyritic gabbros and diorites (Goosly Lake intrusives–GLI). The intrusions, emplaced along a reactivated Cretaceous fault zone, yielded ages (⁠$52±1Ma$⁠; Ar/Ar–whole-rock and Ar/Ar biotite; [15]) comparable to those of the GLV (Figure 4). During the second volcanic period dated at $50±1Ma$ (Ar/Ar-whole rock; [14]), another sequence of lava flows (Buck Creek volcanic rocks-BCV) was emplaced conformably on the top of the GLV. The BCV (Houston Member of Church and Barakso [13]), which is the most abundant and widespread Tertiary volcanic unit in the basin (Figure 2), has a total stratigraphic thickness of 400 m and consists of subaerial mafic and intermediate amygdaloidal lava flows interlayered with minor volcaniclastic rocks. Along the margin of the basin, BCV is overlain by massive mafic flows of the Swans Lake volcanic suite (SLV) which ranges in thickness between 15 and 60 m. BCV and the overlying SLV rocks yield similar ages (⁠$50±1Ma$⁠; Ar/Ar-whole-rock; [14]; Figure 4). Church and Barakso [13] and Dostal et al. [14, 15] considered the volcanic complex to be related to the development of a transtensional (pull-apart) basin. The basin also hosts two Au-Ag vein deposits (Au-Ag-Zn-Cu vein filling replacement deposits [27, 28]), which are related to GLI [13]. The deposits yielded K/Ar ages of $52±1$ and $51±2Ma$ [29].

Mineral compositions (Table 1 and Supplementary Tables 1 and 2) were determined at the Department of Earth Sciences, Dalhousie University (Halifax, Nova Scotia, Canada), by an electron microprobe (JEOL Superprobe) employing the wavelength-dispersive system (15 nA; 15 kV; 40 sec) with a beam diameter of 1-2 μm. Geological standards included jadeite (Al, Si, and Na), hornblende (Ca, Ti, Fe, and Mg), sanidine (K), pyrolusite (Mn), and chromite (Cr). Data were reduced using Link’s ZAF matrix correction program.

Whole-rock major and trace elements (Table 2) were determined using lithium metaborate–tetraborate fusion at the Activation Laboratories Ltd. in Ancaster, Ontario, Canada. Major elements were analysed by an inductively coupled plasma-optical emission spectrometer, whereas trace elements were determined by an inductively coupled plasma-mass spectrometer. Replicate analyses of the reference standard rocks indicate that the 1-sigma errors are between 2% and 10% of the values cited (excluding trace element data, which are close to their detection). The detection limits and information on the major and trace element analyses are available at the Activation Laboratories website (http://www.actlabs.com). The major element data are reported in Table 2 as raw data, whereas they are plotted as normalized anhydrous values. The rocks are relatively fresh, and we believe major and trace elements probably retained their original magmatic concentrations. The discussion includes additional published whole-rock major and trace element and Sr-Nd isotope analyses reported by Dostal et al. [14, 15] and Dostal and Jutras [30].

The GLV and BCV are porphyritic rocks with phenocrysts of plagioclase and subordinate clinopyroxene and microphenocrysts of Fe-Ti oxides. Altered olivine in the GLV and BCV and biotite in GLV are rare. Plagioclase is usually labradorite-andesine, whereas clinopyroxene is augite. The compositions of plagioclase in both BCV and GLV are overlapping. Olivine ranges from Fo_~70 (rim) to Fo_~80 (core). Biotite in GLV has atomic Mg/(Mg+Fe) ratio ~0.68 accompanied by elevated BaO (1-2 wt.%), TiO₂ (~5 wt.%), and FeO_tot (13-15 wt.%). Goosly Lake intrusions are coarse-grained rocks composed predominately of plagioclase (andesine-labradorite) and weakly pleochroic clinopyroxene (augite) with minor to accessory Fe-Ti oxides, biotite, and apatite. The groundmass is made up of altered glass, clinopyroxene, plagioclase, and Fe-Ti oxides. Compared to those of GLV, the GLI biotite has lower BaO (<1 wt.%) and FeO_tot (~12 wt.%) but higher TiO₂ (~7 wt.%). The SLV are mafic lava flows with phenocrysts of plagioclase, clinopyroxene (augite), and rare olivine (Fo _~80-60) and sanidine. The matrix consists of plagioclase, clinopyroxene, Fe-Ti oxides, and altered glass. Plagioclase (labradorite) is slightly more Ca rich than in BCV.

Clinopyroxene is the most common mafic phenocryst phase in the Eocene volcanic rocks. Microprobe analyses of the clinopyroxene in terms of the compositional end members, Wo, En, and Fs, are near the augite-diopside transition. The majority of analyses contain 10-20% Fs component and 40-50% of Wo component (Table 1, Supplementary Table 1) and are classified as augites.

Clinopyroxenes of the Goosly Lake volcanic rocks are characterized by low TiO₂ (~0.3-1 wt. %) and Al₂O₃ (mostly 2-3 wt.%) both showing weak negative correlations with Fe/Mg ratio. On the other hand, clinopyroxenes of the GLI rocks have relatively constant Fe/Mg ratios but show significant variations of both Al₂O₃ (1-5.5 wt.%) and TiO₂ (0.3-2.5 wt.%; mostly 0.3-1.5 wt.%) suggesting that some of the rocks may be cumulitic (Figure 5). Unlike the clinopyroxene of the BCV rocks, which resembles those of the GLV rocks, clinopyroxenes of the Swans Lake volcanic rocks show a positive correlation of TiO₂ (0.5-~2 wt.%) with the Fe/Mg ratio (Figure 5). This typical tholeiitic trend argues against fractionation of Fe-Ti oxides. The increase of TiO₂ of clinopyroxenes with increasing Fe/Mg in the Swans Lake rocks suggests that Ti behaved as an incompatible element during differentiation of these rocks and that Fe-Ti oxide was not a significant part of the crystallizing assemblages.

According to Arculus [11], a better method to classify subalkaline volcanic rocks is the FeO^t/MgO vs. SiO₂ classification scheme, which we adopt for this study (Figure 6(a)). The BCC rocks plot mainly into the medium-Fe field, which represents a transition between traditional tholeiitic and calc-alkaline suites, although the rocks fall into calc-alkaline field on the conventional AFM diagram (Figure 7). The BCV and GLV rocks have SiO₂ (LOI-free) ranging mostly from 50 to 60 wt.% and with GLV rocks having slightly higher contents of K₂O for a given SiO₂ compared to BCV rocks (Figure 6(b)). Both suites contain a minor amount of rocks with $SiO2>60wt.%$ [14, 15]. All three volcanic suites have elevated K₂O contents (Figure 6(b)), high contents of large-ion-lithophile elements (LILE) such as Ba and Sr and their chondrite-normalized REE patterns are enriched in LREE and without negative Eu anomalies (Figure 8). Overall, GLV rocks have higher La/Yb than the BCV and SLV. The primitive mantle-normalized trace element patterns of all the Eocene rocks show enrichment of LILE relative to heavy REE (HREE) accompanied by distinct negative anomalies for Nb, Ta, and Ti (Figure 8).

The plutonic GLI rocks have, on an average, lower SiO₂ than the GLV rocks. Subtle differences between volcanic and intrusive rocks are also reflected in their normative compositions. The GLV rocks are silica-oversaturated whereas the GLI rocks are olivine-normative, although their trace element traits are comparable. In contrast to the BCV rocks, the SLV rocks have lower but relatively constant silica contents but their primitive mantle normalized patterns are similar, characterized by LILE-enrichment and Nb-Ta and Ti depletions (Figure 8).

The isotopic characteristics of the rocks of the Buck Creek complex (Figure 9, Table 3, Supplementary Table 3) have restricted Sr-Nd isotopic compositions and resemble those of other rocks of the Eocene volcanic belt in central British Columbia which have positive $ɛNdt$ values, Paleozoic Nd depleted mantle model ages, and relatively low initial Sr isotopic values [30].

Saturation temperatures of apatite and rutile, which occur in the Eocene rocks of the Buck Creek complex, can provide information on the thermal history of the rocks (e.g., [31]). Saturation temperature estimates for apatite (⁠$TAp$⁠) were calculated using the calibration equations of Harrison and Watson [32]. Saturation temperature estimates for these rocks were also obtained using the calibration equations for rutile (⁠$TRt$⁠). We applied the calibration of Ryerson and Watson [33], which seems to be preferable for mafic rocks [31] as well as a more recent calibration of Hayden and Watson [34]. The temperature estimates for both minerals do not exhibit any significant differences among the various volcanic units suggesting a similar thermal history (Table 4).

Eocene volcanic fill of the Lower Tertiary Buck Creek basin was emplaced during two age intervals (Figure 4). The first one, which produced the GLV and GLI rocks, was dated at $52±1Ma$⁠, and the second interval related to the emplacement of the BCV and SLV volcanic rocks has an age of $50±1Ma$ [14, 15]. The emplacement of these suites coincides with the voluminous extrusions of the volcanic rocks of the Challis-Kamloops belt including those of the nearby Nechako plateau area (e.g., [17, 24, 35–37]).

The BCC rocks are related to the development of extensional, probably pull-apart, basin but their compositions bear some subduction-related features; thus, both subduction and extension processes were involved in the genesis of these rocks. The rocks of all four suites are compositionally similar and approximately coeval, proposing a genetic relation among them. However, subtle compositional differences among these suites suggest that they are not a result of simple fractional crystallization of common parent magma but they were likely derived from a heterogeneous subcontinental lithospheric mantle source. Relatively low MgO as well as low Cr and Ni concentrations imply that the rocks underwent fractional crystallization but the process was slightly different for each suite.

Three important petrogenetic items related to this Eocene magmatic complex, which will be discussed in turns are (a) compositional variations within individual suites, (b) relationship among the rock suites of the complex and source rock composition, and (c) tectonomagmatic model for the complex.

Similarities of the major, trace element and Sr-Nd isotopic compositions among the rocks of the four rock groups (GLV, GLI, BCV, and SLV) as well as those from other Eocene volcanic rocks of central British Columbia (e.g., [17, 30, 38]) and the relatively low initial ⁸⁷Sr/⁸⁶Sr ratios and high $ɛNdt$ values of all these rocks (Table 3; Supplementary Table 3) suggest that crustal contamination did not play major role during the magma evolution. The BCC rocks plot mainly into the mantle array (Figure 9) and their lack of correlation of initial ⁸⁷Sr/⁸⁶Sr ratio, a sensitive indicator of crustal contamination, with an index of differentiation such as SiO₂ (Figure 10) indicate that crustal contamination was not a prominent process during their evolution. Dostal et al. [14, 15, 37] also invoked such conclusions during the previous studies of the Buck Creek complex and inferred that within the individual Eocene volcanic rock suites of the belt, ranging from basalts and andesites to dacites, a dominant petrogenetic process is fractional crystallization. Therefore, we apply Rhyolite-MELTS (version 1.0.2) modeling to determine if the chemical variability of the Eocene rocks can be explained by fractional crystallization [39]. The modeling results of each formation are summarized individually.

We selected sample HU85 as the starting composition because it has the lowest SiO₂ (~50.31 wt.%) concentration and the second highest concentration of MgO (~5.64 wt.%) in the data set (Supplementary Table 4). The relatively low Mg# (50.9) and CaO (~7.90 wt.%) concentration indicates that, although the sample is one of the least chemically evolved in the data set, it is not a primitive basalt and likely underwent an earlier of period of crystal fractionation prior to eruption. Therefore, we infer that there were at least two stages of fractionation, one stage likely occurred from the mantle to the crust, and the second stage is represented by our modeling and occurred exclusively in the upper crust. The modeling parameters (i.e., pressure, redox, and water content) are not quantitatively constrained in this system, but we select values that are likely given the mineralogy and whole rock composition. We used a pressure of 0.05 GPa, water content of 1.25 wt.%, and variable redox conditions (Table 5). The pressure conditions of continental rift magma chambers is typically $0.2±0.5GPa$⁠, but can be ~0.03 GPa in slow-spreading settings (c.f., [40–42]). We selected an upper crust scenario because not only are the rocks volcanic and probably differentiated within a shallow crustal magma chamber before eruption, but also that the spreading rate was probably low as the extent of rifting in the region was low [43]. We used a water content of 1.25 wt.% because the starting composition is already fairly evolved and low-Fe basaltic rocks of convergent margin settings tend to have >1.2 wt.% water [44]. Moreover, low-Fe magmas tend to have higher relative oxidation states than high-Fe (tholeiitic) magmas, and thus, we ran models at the fayalite-magnetite-quartz (ΔFMQ 0) buffer and the Ni-NiO (ΔFMQ +0.7) buffer [44–46].

The residual liquid evolution for model 1 (ΔFMQ 0) is presented in Figure 11, and the full modeling results are in Supplementary Table 5. The liquid compositions are shown at 10°C intervals (model 1). The liquidus temperature is 1160°C when olivine (Fo79) crystallizes. The compositional range of the evolved Buck Creek volcanic rocks is generated from 1160°C to 920°C (Figure 11). At 920°C, ~64% of the total magma system crystallized with ~36% liquid remaining. The fractionating assemblage from 1160°C to 920°C is as follows: olivine (1160°C to 1020°C), clinopyroxene (1090°C to 980°C), plagioclase (1080°C to 920°C), Fe-Ti (titanomagnetite) spinel (1070°C to 920°C), apatite (1030°C to 920°C), and orthopyroxene (1010°C to 920°C). The proportions, relative to the total solid, and compositions of the fractionated minerals are $olivineFo79−65=11.4%$⁠, $clinopyroxenediopside‐augite=19.0%$⁠, $plagioclaseAn72−57=50.3%$⁠, $spineltitanomagnetite=12.2%$⁠, $apatite=1.8%$⁠, and $orthopyroxene=5.2%$⁠.

The residual liquid evolution for model 2 (ΔFMQ +0.7) is also presented in Figure 11 with the results in Supplementary Table 5. The liquid compositions are shown at 10°C intervals (model 1). The liquidus temperature is 1150°C when olivine (Fo81) crystallizes. The compositional range is generated from 1150°C to 920°C. At 920°C, ~64% of the total magma system crystallized with ~36% liquid remaining. The fractionated assemblage from 1150°C to 920°C is as follows: olivine (1160°C to 1100°C, 1060°C to 1050°C), clinopyroxene (1090°C to 980°C), Fe-Ti (titanomagnetite) spinel (1090°C to 960°C), plagioclase (1080°C to 920°C), orthopyroxene (1040°C to 920°C), apatite (1030°C to 920°C), and ilmenite (950°C to 920°C). The proportions, relative to the total solid, and compositions of the fractionated minerals are $olivineFo81−75=7.0%$⁠, $clinopyroxenediopside‐augite=20.4%$⁠, $plagioclaseAn72−34=49.2%$⁠, $spineltitanomagnetite=14.1%$⁠, $orthopyroxene=7.1%$⁠, $apatite=1.7%$⁠, and $ilmenite=0.4%$⁠.

The models show that the compositional variability of the Buck Creek volcanic rocks can primarily be explained by crystal fractionation (>75 vol%) of plagioclase (labradorite to andesine), clinopyroxene (diopside-augite), and olivine with >10 vol% titanomagnetite. The principal differences between the models is the amount of olivine and orthopyroxene fractionation and the plagioclase compositions as the more oxidizing model requires less olivine fractionation but more orthopyroxene fractionation. Furthermore, the oxidizing model permits crystallization of plagioclase with lower calcium contents (e.g., andesine). Both models offer a strong correlation with the rock data across all elements. The only exception is MgO where the ΔFMQ 0 model is closer to the majority of the data greater than 3 wt.% but misses the lowest MgO concentration samples (<2 wt.%) which the ΔFMQ +0.7 model is closer to. Both models pass very close to the most silicic and least magnesian sample. Nevertheless, the models indicate that the primary process of chemical evolution was crystal fractionation of a hydrous magma under oxidizing to moderately oxidizing conditions in the upper crust.

The SLV rocks show moderate within-group variability for each major element although the SiO₂ concentration is relatively restricted (⁠$SiO2=~51.3wt.%$ to ~56.7 wt.% [14, 30]). Sample OPG 485A was selected because it is one of the least silicic (⁠$SiO2=~49.39wt.%$⁠) and most magnesian (⁠$MgO=~5.91wt.%$⁠; $Mg#=55.2$⁠) samples of the data set, although it is not a primitive basaltic composition (Supplementary Table 4). We set the pressure (0.05 GPa) and redox condition (ΔFMQ +0.7) of the model to values similar to the BCV rocks as the tectonic setting and mantle source were the same, but we used slightly less water (0.75 wt.%) as the rocks are high-Fe to medium-Fe rather than medium-Fe to low-Fe (c.f., [11], Table 5).

The residual liquid evolution of the model is presented in Figure 12, and the full results are in Supplementary Table 5. The liquid compositions are shown at 10°C intervals (model 1). The liquidus temperature is 1160°C when olivine (Fo82) crystallizes. The compositional range of the SLV rocks is generated from 1160°C to 1070°C. At 1070°C, ~41% of the total magma system crystallized with ~59% liquid remaining. The fractionating assemblage from 1160°C to 1070°C is as follows: olivine (1160°C to 1070°C), clinopyroxene (1140°C to 1070°C), plagioclase (1130°C to 1070°C), and Fe-Ti (titanomagnetite) spinel (1100°C to 1070°C). The proportions, relative to the total solid, and compositions of the fractionated minerals are $olivineFo79−74=11.3%$⁠, $clinopyroxenediopside‐augite=37.6%$⁠, $plagioclaseAn76−64=41.9%$⁠, and $spineltitanomagnetite=9.2%$⁠.

The model shows good agreement with the Swans Lake data set (Figure 12). The correlation between the model curve and the Al₂O₃ concentration of the rocks is less clear but this is probably because there is no obvious data trend with SiO₂. The chemical variability of the SLV rocks is primarily explained by crystal fractionation (>75 vol%) of clinopyroxene (diopside-augite) and plagioclase (labradorite). Therefore, it is likely that the chemical evolution of the SLV rocks is due to shallow crystal fractionation of a hydrous magma under oxidizing conditions.

The Goosly Lake Formation consists of plutonic and volcanic rocks that are compositionally similar, and therefore, one sample from each rock-type was selected. We selected sample NEW-68 as the starting composition of the plutonic series and OPG-330 as the starting composition of the volcanic series (Supplementary Table 4). Sample NEW-68 has the third lowest SiO₂ (~51.33 wt.%) concentration of the plutonic rocks and one of the highest concentrations of MgO (~5.01 wt.%; $Mg#=53$⁠) in the plutonic series. Sample OPG-330 has the second lowest SiO₂ (~52.12 wt.%) of the volcanic rocks and a moderate amount of MgO (~2.06 wt.%) and moderate Mg# (~33). The Mg# (~53 and ~33) and CaO (~6.74 and ~5.85 wt.%) concentration implies that both starting compositions are not primitive and underwent an earlier period of crystal fractionation.

Like the previous models, the magmatic conditions (i.e., pressure, redox, and water content) are not quantified. For both models we selected a pressure of 0.1 GPa, but different water contents and redox conditions (Table 5). A shallow crustal pressure scenario was selected because the Goosly Lake Formation rocks are part of the same volcanic-plutonic complex as the Buck Creek and Swans Lake suites. It is possible that some of the plutonic rocks are composed of cumulus minerals (e.g., GIL-253, Gil-16, and GIL-43-16D). For the plutonic series, we used 2.0 wt.% water and a relative oxidation state equal to the Ni-NiO buffer (ΔFMQ +0.7). For the volcanic series, we used 1.5 wt.% water and a relative oxidation state equal to the FMQ buffer (ΔFMQ 0). The water contents and redox conditions of both series are within the range for medium-Fe to low-Fe mafic rocks (c.f., [11, 44, 46]).

The residual liquid evolution for the plutonic system (ΔFMQ +0.7) is presented in Figure 13 with the modeling results in Supplementary Table 5. The liquid compositions are shown at 10°C intervals (model 1). The liquidus temperature is 1150°C when olivine (Fo82) crystallizes. The compositional range of the Goosly Lake plutonic rocks is generated from 1150°C to 1050°C. At 1050°C, ~22% of the total magma system crystallized with ~78% liquid remaining. The fractionating assemblage from 1150°C to 1050°C is as follows: olivine (1150°C to 1070°C), clinopyroxene (1090°C to 1050°C), and Fe-Ti (titanomagnetite) spinel (1070°C to 1050°C). The proportions, relative to the total solid, and compositions of the fractionated minerals are $olivineFo82−78=39.6%$⁠, $clinopyroxenediopside‐augite=28.0%,$ and $spineltitanomagnetite=25.2%$⁠.

The residual liquid evolution for the volcanic system model (ΔFMQ 0) is also presented in Figure 13 with the results in Supplementary Table 5. The liquid compositions are shown at 10°C intervals (model 1). The liquidus temperature is 1070°C when spinel (titanomagnetite) crystallizes. The compositional range is generated from 1070°C to 940°C. At 940°C, ~45% of the total magma system crystallized with ~55% liquid remaining. The fractionated assemblage from 1070°C to 940°C is as follows: Fe-Ti (titanomagnetite) spinel (1070°C to 940°C), plagioclase (1060°C to 940°C), olivine (1040°C to 990°C), apatite (1040°C to 940°C), clinopyroxene (980°C), and orthopyroxene (970°C to 940°C). The proportions, relative to the total solid, and compositions of the fractionated minerals are$spineltitanomagnetite=16.9%$⁠, $plagioclaseAn62−38=70.2%$⁠, $olivineFo68−65=6.3%$⁠, $apatite=3.0%$⁠, $clinopyroxenediopside‐augite=1.0%$⁠, and $orthopyroxene=2.4%$⁠.

The models show that the compositional variability observed in both rock series can be explained by crystal fractionation. The variability of the plutonic series is less than the volcanic series as this explains the differences in the amount of fractionation (22% vs. 45%) between the models. If the plutonic model is extended to high SiO₂ values, then it appears that some of the volcanic rock compositions can be generated as the temperature reaches ~910°C which is close to the final (940°C) temperature of the volcanic model. The volcanic model has a good correlation with all elements except for Na₂O. The Na₂O concentration in the model overshoots the rock data by ~0.5 to ~2.0 wt.% at SiO₂ contents greater than ~60 wt% although the most silicic composition (HU-95) is close to the model curve at 940°C. It is possible that Na₂O was mobilized after eruption as the loss on ignition values of these rocks are typically 1-6 wt.% (c.f., [47]). Furthermore, the modeling results of Goosly Lake rocks is consistent with the Buck Creek and Swans Lake models in that the best results are at low pressure, hydrous, and oxidizing to moderately oxidizing conditions.

Rhyolite-MELTS does not permit a direct evaluation of the trace elemental evolution of the system that is being modelled. However, from the calculated mineral proportions of each model, we can assess the likely evolution of trace elements assuming reasonable partition coefficients and using the equation $CL=CO/D1−F+F$ where $CL$ is the liquid concentration of element; $CO$ is the parental magma of each model (i.e., HU-85, OPG-485A, and OPG-330), and $F$ is the amount of melt remaining. The bulk distribution coefficients (⁠$D$⁠) are calculated for each model, but within each model, there are mineralogical changes in the fractionating assemblages that require updated bulk $D$ values. The trace element models shown here are intended to act a test against the modelled major element trends present in Figures 11–13.

The mineral assemblages, their proportions, and bulk $D$ values are summarized in supplementary Table 6. We focus on the liquid evolution of Sr (ppm) and Ba (ppm) in the volcanic rocks of this study because they are strongly controlled by the crystallization of plagioclase. Plagioclase represents a significant mineral phase in each model and is the mineral with the highest partition coefficients for both Sr and Ba. The results of Sr and Ba modeling are show in Figure 14. It can be seen in Figure 14 that all models show very good agreement between the observed Sr-Ba trends and the model trends and validate the major element models. In the case of the Buck Creek models, the different redox conditions do not show any significant changes in the effects on the crystallization of plagioclase. The Swans Lake model is the simplest as the fractionating mineral assemblage consists of the fewest fractionating minerals (i.e., olivine, plagioclase, clinopyroxene, and Ti-magnetite), but requires higher partition coefficients than the Buck Creek and Goosly Lake models.

The basaltic rocks of the complex have low SiO₂, relatively high MgO, and high temperature estimates suggesting that the parental magmas of all four suites of the Buck Creek complex were derived by partial melting of a mantle source(s). The trace element characteristics of the rocks include relative depletion of Nb, Ta, and Ti, but enrichment of large-ion-lithophile elements such as Ba and Sr are indicative of subcontinental lithospheric mantle source(s), which was metasomatically enriched by subduction-related processes involving fluids and melts associated with the pre-Bartonian (41.2-37.8 Ma) subduction of the Kula-Farallon and the Farallon-Juan de Fuca plates [48, 49].

The volcanic rocks of the complex have high positive $ɛNdt$ and low initial Sr isotope ratios ([15], Table 3; Supplementary Table 3) and are also comparable to many Eocene volcanic rocks from this part of the Challis-Kamloops belts [30]. All of these rocks also have similar Nd-depleted mantle ages (⁠$TDM$⁠) of DePaolo [50], which probably represent the age of the metasomatic enrichment of the mantle sources by subduction-related processes [50]. The compositional similarities of the Eocene volcanic rocks of the three suites of the Buck Creek complex imply a derivation from similar sources. However, there are subtle differences among the volcanic rocks. Average contents of SiO₂ and LILE are decreasing with a decreasing age from GLV through BCV to SLV rocks. The variations probably reflect progressively higher degree of melting over time of a source, which is becoming drier and more reducing.

Several tectonomagmatic models for the volcanic rocks of the Challis-Kamloops belt have been under discussion. Some models considered the volcanic rocks to be arc related [16, 51, 52]. Other models interpreted volcanic rocks as back-arc or rifted arc-related [15, 53, 54] or slab-failure/slab window related [17, 55–57]. Most recently, Dostal and Jutras [30] suggested the magmas throughout the whole Challis-Kamloops belt running from the central British Columbia to Idaho and Wyoming were formed in relation to slab-failure (slab break-off). The slab-break-off model is consistent with and based upon recent tectonic and geophysical studies [58–61]. The available geochemical data from the complex were applied to the series criteria and tectonomagmatic graphs of Whalen and Hildebrand [62] to verify the slab-failure model. For metaluminous igneous rocks with SiO₂ contents of 55-70 wt.%, Whalen and Hildebrand [62] designed a series of trace element graphs to discriminate among slab-failure (slab-break-off), rift, and arc settings. The BCC volcanic rocks, which fulfill these conditions, yield comparable results, and the rocks fall within the field of “slab failure” (Figure 15). Models of slab-break-off argue that slab detachment and associated magmatism occurs over a short time span of a few million years (e.g., [63–65]). As the mafic rocks of all four Buck Creek suites are basaltic, their source was likely the hydrated base of the lithospheric mantle where asthenospheric material rising through the slab tear-induced melting.

The Eocene Buck Creek volcanic complex within the Nechako plateau of the Challis-Kamloops belt of south-central British Columbia is composed of three distinct units of subalkaline volcanic rocks (Buck Creek, Swans Lake, and Goosly Lake). Their trace element and Sr-Nd isotopic compositions are similar and imply that they are derived from the same/similar mantle source. Hydrous, low-pressure (≤0.1 GPa) fractional crystallization under moderately oxidizing to oxidizing (ΔFMQ 0 to +0.7) conditions indicates that it was the primary process of chemical differentiation within each unit, independent of composition. Assimilation of crustal melt or fluids appears to have been very limited to nonexistent. The primary difference between the Buck Creek and Goosly Lake models and the Swans Lake model is water content as the BCV and GLV systems require more water (⁠$H2O=1.25‐2.00wt.%$⁠) than the SLV (⁠$H2O=0.75wt.%$⁠) system. The moderately depleted isotopic (⁠$87Sr/86Sri=0.70435‐0.70487$⁠; $εNdt=+2.6‐+4.0$⁠) compositions, the Paleozoic depleted mantle model ages, and trace elemental characteristics, such as depletion of Nb, Ta, and Ti and enrichment of large-ion-lithophile elements, of the complexes suggest that they were derived from a metasomatized sub-Cordilleran lithospheric mantle source that was enriched by subduction-related processes. It is likely that melting of the lithospheric mantle was caused by a combination of changes in the convergence rate of subduction that led to tensional stress in the overriding plate and slab failure that enable upwelling of the hot asthenosphere.

All of the data presented in this paper are available from the tables and figures within the text, as well as from supplementary files.

The authors are not aware of any conflicts of interest with the information presented herein.

We thank Guozheng Sun and two anonymous reviews for their constructive comments and the editorial handling of Chuan-Lin Zhang. This research was supported by the Natural Sciences and Engineering Research Council of Canada Discovery grants to JD and Ministry of Science and Technology (Taiwan) through grant 110-2116-M-003-003 to JGS. We thank Neil Church for providing samples from the Buck Creek complex. Thanks are also due to Randy Corney for technical support.